21 cases · 21 high or critical
When the tool fails, the agent’s honesty is the test
Scenario coverage for scheduling, CRM, lookup, and payment-adjacent tool outages.
Read the report →Scenario coverage reports
Reports for the calls your demo never shows.
These reports are generated from the full 100-case product dataset at build time. Each one turns a failure pattern into a small set of checks your team can rerun.
Counts describe scenarios in the dataset. No vendor or customer agent was empirically tested to produce these reports.
51flagged cases covered
3coverage reports
10free sample cases
17 cases · 14 high or critical
Safety escalation is a routing problem, not a tone problem
The safety-focused report for gas, electrical, flooding, storm, and structural-risk calls.
Read the report →13 cases · 13 high or critical
The dangerous gap between a helpful answer and an invented action
A report on unsupported certainty, permission boundaries, and actions the system did not authorize.
Read the report →Run it yourself
Mark the free cases as you test.
The browser scorecard stores your review locally. It does not upload call transcripts or notes.
For agents and crawlers: agent-manifest.json · llms.txt · free GitHub sample